Jie He

Université de Sherbrooke, Canada

Jaime de Melo*

FERDI and university of Geneva, Switzerland

Haisheng Yang

Sun Yat-sen University, China

Shanghai, May 25,2015

Trade in virtual carbon:

Evidence from spatial econometric models

* Ferdi, Fudan University, and Paris 1 University joint workshop

“China and Globalization”

Outline

Pollution Havens & Virtual Trade in

Carbon: stylized facts

Pollution havens effects: small

Virtual Trade in Carbon: large

discrepancy in results

Framework

Data and estimation

(Preliminary) Results

Pollution Havens and Virtual Trade

in Carbon: Stylized facts

Pollution Havens: Stylized Facts

Evidence mostly from SO2—a local pollutant

Energy-intensive sectors are weight-reducing = Not

footlose (not much world-wide displacement of production for SO2-intensive sectors over period 1990-2000--see extra slides) Relevant for CO2?

Global studies: Small PH effects in bilateral trade (but strong

composition effects as NN dominates NS trade ver 1990-00 so Pollution Content of Imports (PCI) is not much affected by environment policies-next slide)

Factoring in FDI--mostly directed to EPZs likely to lead to cleaner exports (Dean and Lovely (2009) for China).

-10

-50

51

0

Biochemical oxygen demandCO in air

Fine particulates in airNO2 in air

SO2 in airTotal suspended particulates

Total suspended solids in waterToxic metal pollution

Toxic pollutionVolatile organic compounds

PCI Decomposition for 10 major pollutants, in (%)

fe ph tot

TOT PCIkij = 1(‘fundamental’)+ TOT= (1+ PH)(1 + FE)

CO2 emissions from fuel Combustion by region (1)

Source: Victor (2015, figure 1 from IPPCC (SPM) WGIII)

SUM≈32Gt

Bilateral Product-level embodied emissions in Trade (estimates for 2006 using average world emission factors)

7

Absolute volumes

EEI: Emissions

Embodied in Imports

EEE: Emissions

Embodied in Exports

5/1080 products

account for 15%

of emissions.;

10% account for

50% of emissions

Aggregating gives 2006 patterns : ‘Production centers’ (Indonesia, Australia);

‘Consumption centers’ (Singapore, UK);

‘production and consumption centres’: (US, FR) import ‘downstream’ products while

Italy and Germany export ‘downstream’ products.

Annex B- non-annex B groupings makes little sense.

Do patterns evolve over time?

Large differences across

studies for the SAME year

4.4Gt <EET (2004)< 6.2Gt

Problems:

(1) Estimates typically for

1 year

(2) lack of mechanisms to

account for the emissions

produced in one country

and consumed in another

Our approach:

‘Augmented’ EKC

estimated using spatial

econometric methods

Embodied carbon in

China’s trade in 2004

Framework

Reduced form representation of CO2 emission

Omitting country and time subscripts, the standard EKC model in panel relates

CO2 emissions, E, to a vector, X, of country-specific variables

𝐸 = 𝑋𝛽 + 𝜇 + 𝛾 + 𝜖 (1)

Where:

E: CO2 emissions from fossil fuels in production (territorial)

X : a vector of country-specific variables (income, environmental policy

stringency, population density)

γ : a dummy variable that controls for country-specific time-invariant omitted

variables

μ: a dummy variable that captures common time-specific shocks

This specification has been estimated many times under the strong

identification assumption that the condition for pooling countries is satisfied

(See Ordas (2008))

Augmented representation

Trade between countries lead to trade in virtual carbon (like trade in virtual water)

Emissions in country i measured at the consumption (rather than production)

level also depend on the emissions of its trade partners j≠i

𝐸𝑖𝑡 = 𝑋𝛽 + 𝜆𝐸𝑗≠𝑖,𝑗 + 𝜇 + 𝛾 + 𝜖 (2)

Where:

λ: captures the interdependency in emissions or «connectivity» between i and

other countries (after controlling for differences in environmental policies)

Larger volume of bilateral trade signifies stronger economic interdependence

Here trade between countries is not modelled. It depends on

Trade costs

Environmental policies

Political regimes

Common language

Common culture/religions, etc.

Case I: Symmetric countries, 1 sector

• Emissions in production for countries i and j is given by

𝐸𝑖𝑃 = 𝑌𝑖𝑖𝑎 𝜃𝑖 and 𝐸𝑗

𝑃 = 𝑌𝑗𝑗𝑎 𝜃𝑗 (3)

• Where 𝑘𝑎 𝜃𝑘 , k = i, j is emission intensity depending on strigency of environtmental policy (Brock and Taylor, 2008)

• From national accounts:

Y≡C+X-M

Y : GDP, C : Domestic consumption, X : exports, M : imports

• Virtual carbon in bilateral trade can be written as:

𝐸𝑖𝐶 = 𝑌𝑖𝑖𝑎 𝜃𝑖 − (𝑀𝑖−𝑋𝑖)𝑖𝑎 𝜃𝑖 (4)

𝐸𝑗𝐶 = 𝑌𝑗𝑗𝑎 𝜃𝑗 − (𝑋𝑗−𝑀𝑗)𝑗𝑎 𝜃𝑗 (5)

• Since (𝑋𝑖−𝑀𝑖) = (𝑀𝑗−𝑋𝑗), manipulations relate emissions in consumption to intensities in

production and to patterns of trade :

𝐸𝑖𝑃 = −

𝑖𝑎 𝜃𝑖

𝑗𝑎 𝜃𝑗

𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑒𝑚𝑖𝑠𝑠𝑖𝑜𝑛 𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦

𝑀𝑖−𝑋𝑖𝑌𝑗

𝑒𝑙𝑒𝑚𝑒𝑛𝑡𝑠 𝑜𝑓 𝑡𝑟𝑎𝑑𝑒 𝑚𝑎𝑡𝑟𝑖𝑥

1+𝑋𝑖−𝑀𝑖

𝑌𝑖

𝑠𝑝𝑎𝑡𝑖𝑎𝑙 𝑚𝑎𝑡𝑟𝑖𝑥

𝐸𝑗𝑃 +

1

1+𝑋𝑖−𝑀𝑖

𝑌𝑖

𝐸𝑖𝑃

𝐸𝐾𝐶 𝑟𝑒𝑠𝑐𝑎𝑙𝑒𝑑 𝑏𝑦 𝑡𝑟𝑎𝑑𝑒 𝑟𝑎𝑡𝑖𝑜

(6)

Estimation function can be written as:

𝐸 = 𝑊𝐸 + 𝑋𝛽 + 𝜇 + 𝛾 + 𝜖 (7) Where W is a N N square weight matrix, whose elements measure the

share of the net import by country i from country j (illustration for the case

of three countries)

W=

0 𝑇𝑟𝑎𝑑𝑒𝑡1←2 𝑇𝑟𝑎𝑑𝑒𝑡

1←3

𝑇𝑟𝑎𝑑𝑒𝑡2←1 0 𝑇𝑟𝑎𝑑𝑒𝑡

2←3

𝑇𝑟𝑎𝑑𝑒𝑡3←1 𝑇𝑟𝑎𝑑𝑒𝑡

3←2 0

(8)

𝑇𝑟𝑎𝑑𝑒𝑡𝑖←𝑗 =

𝑖𝑡

𝑗𝑡

𝑖𝑚𝑝𝑜𝑟𝑡𝑡𝑖←𝑗−𝑒𝑥𝑝𝑜𝑟𝑡𝑡

𝑖→𝑗

𝑌𝑗𝑡 (9)

(the arrows show the direction of the merchandise movements between countries)

Case I: Symmetric countries, 1 sector (end)

• To take into account the differences in the impacts on virtual carbon between : – Trade of carbon intensive products (h)

– Trade of other products (l)

𝐸𝑖𝑃 = −

𝑖𝑎 𝜃𝑖

𝑗𝑎 𝜃𝑗

×

(𝑀𝑖ℎ−𝑋𝑖

ℎ)𝑗ℎ𝑎 𝜃𝑗

ℎ

𝑌𝑗𝑗𝑎 𝜃𝑗

+(𝑀𝑖

𝑙−𝑋𝑖𝑙)𝑗

𝑙𝑎 𝜃𝑗𝑙

𝑌𝑗𝑗𝑎 𝜃𝑗

𝑡𝑟𝑎𝑑𝑒 𝑚𝑎𝑡𝑟𝑖𝑥

1 +(𝑋𝑖

ℎ−𝑀𝑖ℎ)

𝑌𝑖×𝑖

ℎ𝑎 𝜃𝑖ℎ

𝑖𝑎 𝜃𝑖

+(𝑋𝑖

𝑙−𝑀𝑖𝑙)

𝑌𝑖×𝑖

𝑙𝑎 𝜃𝑖𝑙

𝑖𝑎 𝜃𝑖

𝐵𝑖𝑙𝑎𝑡𝑒𝑟𝑎𝑙 𝑡𝑟𝑎𝑑𝑒 𝑚𝑎𝑡𝑟𝑖𝑥

× 𝐸𝑗𝑃 +

𝐸𝑖𝑃

1 +(𝑋𝑖

ℎ−𝑀𝑖ℎ)

𝑌𝑖×𝑖

ℎ𝑎 𝜃𝑖ℎ

𝑖𝑎 𝜃𝑖

+(𝑋𝑖

𝑙−𝑀𝑖𝑙)

𝑌𝑖×𝑖

𝑙𝑎 𝜃𝑖𝑙

𝑖𝑎 𝜃𝑖

𝑆𝑐𝑎𝑙𝑒𝑑 𝐸𝐾𝐶

So the bilateral trade matrix W is now written as:

𝑊 = ℎ𝑊ℎ

𝐻𝑒𝑎𝑣𝑦 𝑝𝑜𝑙𝑙𝑢𝑡𝑖𝑛𝑔 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑠′𝑡𝑟𝑎𝑑𝑒 𝑚𝑎𝑡𝑟𝑖𝑥

+ 𝑙𝑊𝑙

𝑜𝑡ℎ𝑒𝑟 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑠′𝑡𝑟𝑎𝑑𝑒 𝑚𝑎𝑡𝑟𝑖𝑥

Where the element of 𝑊ℎ is(𝑀𝑖

ℎ−𝑋𝑖ℎ)

𝑌𝑗 and the element of 𝑊ℎ is

(𝑀𝑖𝑙−𝑋𝑖

𝑙)

𝑌𝑗

ℎ captures 𝑗

ℎ𝑎 𝜃𝑗ℎ

𝑗𝑎 𝜃𝑗

and 𝑙 captures 𝑗

𝑙 𝑎 𝜃𝑗𝑙

𝑗𝑎 𝜃𝑗

Case II: Symmetric countries, 2 sectors (H and L-carbon)

Emission transfer of i via

net imports from j is

reflected by negative

coefficient

• To distinguish trade between countries according to their environmental policies

– with similar environtmental policies (NorthNorth, SouthSouth)

– with different environmental policies (NorthSouth, SouthNorth)

Trade matrix now divided into four parts

W= 𝑊𝑁𝑁 𝑊𝑁𝑆

𝑊𝑆𝑁 𝑊𝑆𝑆= NN 𝑊𝑁𝑁 0

0 0+NS

0 𝑊𝑁𝑆

0 0+SN

0 0𝑊𝑆𝑁 0

+SS0 00 𝑊𝑆𝑆

• Further distinction of trade according to carbon-intensity category: – Trade in carbon intensive products (h)

– Trade in other products (l)

𝑊 = 𝑁𝑁ℎ 𝑊𝑁𝑁

ℎ +𝑁𝑆ℎ 𝑊𝑁𝑆

ℎ +𝑆𝑁ℎ 𝑊𝑆𝑁

ℎ + 𝑆𝑆ℎ 𝑊𝑆𝑆

ℎ

𝐻𝑒𝑎𝑣𝑦 𝑝𝑜𝑙𝑙𝑢𝑡𝑖𝑛𝑔 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑠′𝑡𝑟𝑎𝑑𝑒 𝑚𝑎𝑡𝑟𝑖𝑥

+ 𝑁𝑁𝑙 𝑊𝑁𝑁

𝑙 +𝑁𝑆𝑙 𝑊𝑁𝑆

𝑙 +𝑆𝑁𝑙 𝑊𝑆𝑁

𝑙 + 𝑆𝑆𝑙 𝑊𝑆𝑆

𝑙

𝑜𝑡ℎ𝑒𝑟 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑠′𝑡𝑟𝑎𝑑𝑒 𝑚𝑎𝑡𝑟𝑖𝑥

Case III: 2-country groups, 2 sectors (H and L-carbon)

Data and estimation

Data and estimation Data: 1996-2009 (14 years), 56 countries (see coverage in next slide)

1. Bilateral trade data from UNcomtrade database (mirro data of export used)

2. CO2 emission from fossil fuel combustion (WDI)

3. Other macroeconomic data from WDI and other sources

Econometric Strategy:

1. 2SLS proposed by Kelejian and Prucha (1998) to take care of the endogeneity of Ej

2. GMM to take care of the heteroskedasticity of panel data (Lee, 2003)

0

5

10

15

20

25

30

35

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

Gt

Year

The CO2 emissions included in our study

World CO2 emission (WDI)

Emission included in our study with 56 countries

Preliminary Results:

Symmetric countries, one product

category

Ei=Ln(CO2) Simple EKC Spatial weight matrix adjusted by

carbon efficiency

FE RE FE RE

W*Ej -0.0027 -0.0029**

(2.19)** (2.32)

LGDPPC 1.64*** 1.80*** 1.56*** 1.71***

(10.13) (11.21)*** (9.35) (10.35)

LGDPPC2 -0.06*** -0.07*** -0.05*** 0.07***

(5.43) (6.96) (5.03) (6.45)

LPOPDEN 1.69*** 1.36*** 1.69*** 1.36***

(16.58) (15.48) (16.63) (15.54)

LER -0.17*** -0.19*** -0.16*** -0.19***

(5.70) (6.66) (5.69) (6.65)

C -4.32*** -3.06*** -3.88*** -2.60***

(5.92) (4.34) (5.15) (3.57)

R2 0.9980 0.5662 0.9979 0.5589

Country Effect Yes Yes Yes Yes

Year effect Yes Yes Yes Yes

Hausman 44.25 45.58

Table 1. total trade spatial weight matrix

2.0

0e+

07

2.2

0e+

07

2.4

0e+

07

2.6

0e+

07

Wo

rld

CO

2 e

mis

sio

ns K

ton

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009Year

co2_world co2_world_predict

co2_world_trade_removed

World CO2 emission observed vs. predicted and trade impact

CO2 predicted with

trade’s impact

removed

share of CO2

accounted for by

trade

CO2 predicted by

the whole model

-.5

0.5

11.5

2

% o

f w

orl

d v

irtu

al carb

on e

mis

sio

ns

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009year

% of world virtual carbon variation due to trade between 56 countries

Before 2006, trade

led CO2 emission to

reduce (composition)

After 2006, trade led

CO2 emission to

increase (participation

of China)

-600000

-400000

-200000

0

200000

400000

600000

800000

1000000

1200000

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

CO

2 e

mis

sio

n (

Kto

n)

Year

Decomposition of the emission transferred via trade between 56 countries

USA TUR

SWE SVN

SVK PRT

POL NZL

NOR NLD

MEX KOR

JPN ITA

ISR ISL

IRL HUN

GRC GBR

FRA FIN

EST ESP

DNK DEU

CZE CHL

CHE CAN

AUT AUS

ZAF URY

UKR TUN

THA RUS

PRY PHL

PER PAK

MYS MAR

LTU KEN

IND IDN

HRV EGY

ECU DZA

CHN BRA

BOL BGR

total

US

China

Germany

Russia

Preliminary Results

Symmetric countries, 2 sectors (H and L-carbon)

Spatial weight matrix adjusted by

carbon efficiency

FE RE

H Carbon trade -0.0025*** -0.0028**

-2.9252 -2.6025 L Carbon trade -0.0043 -0.0033

-1.2277 -0.8998 Ln(GDP per capital) 1.5720*** 1.7098***

7.6587 7.5625 Ln(GDP per capital)^2 -0.0549*** -0.0683***

-4.2676 -5.0561 Ln(People density) 1.6900*** 1.3545***

12.1460 11.3636

Ln(environmental regulatory

intensity) -0.1654*** -0.1902***

-4.1361 -4.6971 Constant -3.9507*** -2.6106***

-4.5901 -2.7473 R2 0.9978 0.5580

Table 3. Results with H- carbon and L-carbon sectors distinguished

-.5

0.5

11

.52

1995 2000 2005 2010time

co2_trade_c_non_carbon_world_p co2_trade_c_carbon_world_p

% CO2 transferred via trade carbon vs non carbone leakage

CO2 variation

due to H carbon

trade

CO2 variation

due to L

carbon trade

-600000

-400000

-200000

0

200000

400000

600000

800000

1000000

1200000

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

Kto

ns

Year

Decomposition of variation of CO2 via trade in carbon

leakage products

OECD

non-OECD

Total

Preliminary Results

2- country groupings (N,S) and two product categories (H,L-carbon intensity)

Spatial weight matrix adjusted by carbon

efficiency

FE RE

H Carbon Non OECD inside trade 0.0346 0.0353

1.62 1.18

H Carbon Import by Non OECD from OECD -0.0497** -0.0680**

-2.24 -2.27

H Carbon Import by OECD from Non OECD 0.0096** -0.0156**

2.26 -2.02

H Carbon OECD inside trade -0.0171** 0.0435**

-2.53 2.35

L Carbon Non OECD inside trade 0.0164 -0.0233

0.39 -0.64

L Carbon Import from OECD to Non OECD -0.0395 -0.0377

-1.20 -0.91

L Carbon Import from Non OECD to OECD 0.0050 0.0242

0.40 0.85

L Carbon OECD inside trade 0.0161 -0.0268

1.54 -1.34

Ln(GDP per capital) 1.72*** 1.69***

7.53 13.22

Ln(GDP per capital)^2 -0.06*** -0.07***

-4.81 -10.58

Ln(People density) 1.63*** 1.05***

11.40 21.95

Ln(environmental regulatory intensity) -0.16*** -0.22***

-4.03 -18.85

Constant -4.23*** -0.74

-4.18 -1.28

R2 0.9979 0.4933

Table 4. results with trade block and H-L carbon trade distinguished

World CO2 emission observed vs predicted by model with 8 trade

matrix (table 4, fixed effect)

2.0

0e+

072.2

0e+

072.4

0e+

072.6

0e+

072.8

0e+

07

Wo

rld

CO

2 e

mis

sio

ns K

ton

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009Year

co2_world CO2_predicted_world_NCNS

CO2_predicted_NCNS_wotrade

World CO2 emission observed vs. predicted and trade impact

CO2 predicted with

trade’s impact

removed

CO2 predicted by

the whole model

CO2 emission

attributable

to trade (H+L)

-1500000

-1000000

-500000

0

500000

1000000

1500000

2000000

2500000

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

Kto

n

Decomposition of CO2 variation via trade into country and

product types

Non Carbon_OECD

Carbon-OECD

Non Carbon-non_OECd

Carbon-non_OECD

Total

Summary so far

• Carbon is exported by non OECD countries towards OECD

countries

• Increases in virtual trade in carbon due to participation in trade

of countries with lower carbon emission efficiency

• Role of trade on aggregate carbon emissions increases with time

• Some evidence of pollution havens especially in carbon leakage

risk products’ trade activities.

• Trade in carbon intensive products is responsible for most of the

virtual carbon increase caused by trade.

• Incorporating product-level carbon intensity (Sato 2014) will give

more accurate results than OECD (arbitrary) classification of 84

carbon intensive products.

• Simple OECD /Non OECD is misleading when discussing

environmental trade policy for carbon intensive activities.

References

• Grether, J.M., N. Mathys, J. de Melo (2010) ), “Is Trade Bad for the Environment?

Decomposing world-wide SO2 Emissions”, Review of World Economics,vol. 145(4), 713-29.

• Grether, J.M., N. Mathys, J. de Melo (2012) « Unravelling the World Wide Pollution Haven

Effect », Journal of International Trade and Development, 21(1), 131-62

• Kelejian, H. and T. Prucha (1999) « A Generalized Moments Estimator for the Autoregressive

parameter in a spatial Model » International Economic Review,40, 509-33

• Ordás Criado, C. 2008. "Temporal and Spatial Homogeneity in Air Pollutants Panel EKC

Estimations," Environmental & Resource Economics, vol. 40(2), pages 265-283

• Peters, et al (2011) « CO2 embodied in international Trade with implications for global

climate policy », Proceedings of the National Academy of Sciences

• Sato (2014) « Product Level Carbon Embodied in Bilateral Trade », Ecological Economics,

106-17

• Sato (2014) « Embodied Carbon in Trade: A Survey of the Empirical Literature », Journal of

Economic Surveys, 28(5), 831-61.

• Victor, D. (2015) « Climate Clubs» E-15 Working group

Evolution of carbon content of trade is an indirect (and partial) measure of

carbon leakage as evolution is also dependent on other “fundamental” aspects

comparative advantage that are independent of trade.

2-good 2-country example: North (N) & South (S) produce a clean and a dirty

good with same endowments and same emissions per unit of output for dirty

good.

•With stricter environmental stds in North: North trade will be embodied with

emissions (i.e. PCINS > 0) while : (PCISN = 0) since it imports clean good.

•Globalization via reduction in transport costs or reduction in trade barriers will

lead S to specialize in dirty products according to PH effect. Same results from

environmental stds. tightening in N.

•Now assume that dirty industries are k-intensive. Then FE effect could

determine comparative advantage even with stricter environmental stds. in N.

Trade liberalization would then lead to relocation to dirty industry to N.

•In this framework, PHH holds if PH effect dominates FE effect in absolute

value (see Grether et al. JITD 2012)

Illustrative example of virtual trade in carbon